Fleet's Tau AI cuts copper-grade prediction error at Chile project
Fleet Space Technologies has launched Tau 3.8, an AI world model for mineral systems that the company says predicted copper grades with 38% less error than conventional geostatistical estimates at ATEX Resources' Valeriano project in Chile. The model was trained on ten million synthetic geological worlds and is available in the United States and Canada.
- Fleet launched Tau 3.8 in the United States and Canada.
- Tau was trained on ten million synthetic geological worlds.
- The model cut copper-grade prediction error by 38% at ATEX's Valeriano project, according to Fleet.
- Error was up to 67% lower on the deepest step-out holes.
- Fleet is working with Nvidia on infrastructure for future subsurface models.
AI model targets the copper supply challenge
Fleet says Tau is designed to help mining companies reconstruct hidden geology, rank targets for resource expansion and identify which additional measurement would reduce uncertainty the most. The aim is to improve decisions about where to drill, using drilling, geology, geophysics, mineralogy and Fleet's 3D seismic data in one model.
The company launched Tau as demand for copper grows alongside North America's data-center and power infrastructure build-out. S&P Global forecasts that data-center copper demand will rise from 1.1 million metric tons in 2025 to 2.5 million metric tons by 2040. According to S&P Global, AI training will account for 58% of that demand by 2030.
Fleet says discovery is a major constraint because deposits are slow, expensive and uncertain to find, while drillholes sample only a small part of the ground below them. "North America is building an AI economy on a mineral base it hasn't yet found," Fleet Co-Founder and CEO Flavia Tata Nardini said.
Valeriano test shows lower prediction error
At ATEX Resources' Valeriano copper-gold project in Chile, Tau combined Fleet's ambient noise tomography and magnetotelluric surveys with Comet and machine learning. In a test using drillholes withheld from the model, Fleet says Tau predicted copper grades with 38% less error than a conventional kriging estimate built from the same drilling.
On the deepest step-out holes, the company reports that the error was up to 67% lower. Fleet says the conventional estimate was an independent exercise and was not related to the 2025 SRK Mineral Resource Estimate.
Drilling is typically the largest cost in mineral exploration, and a single deep diamond drillhole can cost hundreds of thousands of dollars, according to the company. Fleet says more accurate grade predictions can help teams prioritize targets, define and expand resources with less drilling, and reduce the number of holes that end in barren ground.
Tau combines synthetic geology with seismic data
Fleet Chief Scientific Officer Dr. Gerrit Olivier said the company created ten million simulated geological environments, modeled what seismic sensors and drillholes would detect in each, and trained Tau to infer the geology between those partial measurements. Unlike a system focused on language, Tau is intended to learn the structure, geometry and composition of physical mineral systems.
Seismic data helps constrain the depth, orientation, geometry and continuity of structures between drillholes. As new information arrives, Tau updates its subsurface model and recommends the next measurement, an approach Fleet calls Agentic Geoscience.
"Tau does not remove the geologist from the decision," Nardini said. "It gives them a quantified view of uncertainty before the drill bit provides the answer." Fleet is working with Nvidia on accelerated computing infrastructure for high-resolution subsurface models, and it is an industrial affiliate of Stanford University's Mineral-X program.
